The problem is that in general "code" can't do that because the code that's switching stack doesn't necessarily conceptually control the stack it's switching to/from. One SS:SP might point to a userspace process, for instance, and the other SS:SP be for the kernel. Both stacks are "live" in that they still have data that will later be context switched back to. You might be able to come up with a convoluted workaround, but since "switch between stacks" was a design goal (and probably assumed by various OS designs they were hoping to get ports for), the race that meant you can't do it safely was definitely a design bug.
